Awesome

ov-mnist

An overhead and satellite analog to the classic MNIST Digit Recognizer dataset. Also include CSV versions for each image to match with other statistical and computer vision methods that extract features or build models without employ pixel-based deep learning.

Consists of training and test splits (9:1) and includes 1000 images per class as 28x28 pixel grayscale versions

Classes include:

class label train_count test_count

car 0 888 112

harbor 1 889 111

helicopter 2 655 82

oil_gas_field 3 888 110

parking_lot 4 888 112

plane 5 889 111

runway_mark 6 801 100

ship 7 889 111

stadium 8 843 105

storage_tank 9 889 111